invoke-ai/InvokeAI · error · ValueError

Wan latents-to-video expects a 5D latent tensor [B, C, T, H,

Error message

Wan latents-to-video expects a 5D latent tensor [B, C, T, H, W]; got {tuple(latents.shape)}.

What it means

After promoting 4D tensors to 5D, invoke() requires the latent tensor to be rank 5 ([B, C, T, H, W]). Anything else (1D, 2D, 3D, or 6D+) cannot be interpreted as video latents, so a ValueError with the actual shape tuple is raised.

Source

Thrown at invokeai/app/invocations/wan_latents_to_video.py:99

    latents: LatentsField = InputField(description=FieldDescriptions.latents, input=Input.Connection)
    vae: VAEField = InputField(description=FieldDescriptions.vae, input=Input.Connection)
    fps: int = InputField(
        default=16,
        ge=1,
        le=120,
        description="Frames-per-second for the encoded MP4. Wan 2.2 was trained at 16 FPS.",
    )

    @torch.no_grad()
    def invoke(self, context: InvocationContext) -> VideoOutput:
        latents = context.tensors.load(self.latents.latents_name)
        _validate_video_latent_batch(latents)
        if latents.ndim == 4:
            # Promote 4D (single-frame) to 5D so this node can also serve as a
            # one-frame "video" encode if someone wires it that way.
            latents = latents.unsqueeze(2)
        if latents.ndim != 5:
            raise ValueError(
                f"Wan latents-to-video expects a 5D latent tensor [B, C, T, H, W]; got {tuple(latents.shape)}."
            )
        if any(size == 0 for size in latents.shape[2:]):
            raise ValueError("Wan latents-to-video requires non-empty temporal and spatial dimensions.")

        vae_info = context.models.load(self.vae.vae)
        if not isinstance(vae_info.model, AutoencoderKLWan):
            raise TypeError(f"Expected AutoencoderKLWan for Wan VAE, got {type(vae_info.model).__name__}.")

        if latents.shape[1] != vae_info.model.config.z_dim:
            raise ValueError(
                f"Latent channel mismatch: these latents have {latents.shape[1]} channels but the "
                f"selected VAE expects {vae_info.model.config.z_dim}. A14B models need the 16-channel Wan 2.1 VAE; "
                "TI2V-5B needs the 48-channel Wan 2.2 VAE."
            )

        _, _, t_lat, h_lat, w_lat = latents.shape
        spatial_scale = getattr(vae_info.model.config, "scale_factor_spatial", None) or 8

View on GitHub (pinned to 0b6a024f2f)

Solutions

Ensure the input is a Wan denoiser output of rank 5 [B, C, T, H, W].
If you have single-frame 4D latents [B,C,H,W], the node auto-promotes them — verify no extra squeeze/dim edits occurred upstream.
Print latents.shape before invoking and reshape correctly (e.g. latents.unsqueeze(0) for a missing batch dim).

Example fix

// before
latents = torch.randn(16, 8, 64)  # 3D, invalid
video = wan_latents_to_video(latents=latents)
// after
latents = torch.randn(1, 16, 8, 64, 64)  # [B, C, T, H, W]
video = wan_latents_to_video(latents=latents)

Defensive patterns

Strategy: validation

Validate before calling

if latents.ndim not in (4, 5):
    raise ValueError(f"Expected 4D or 5D Wan latents, got shape {tuple(latents.shape)}")

Type guard

def is_video_latent_tensor(t) -> bool:
    return isinstance(t, torch.Tensor) and t.ndim in (4, 5)

Try / catch

try:
    video = node.invoke(context)
except ValueError as e:
    if "5D latent tensor" in str(e):
        latents = fix_rank(latents)  # e.g. unsqueeze(0) for missing batch
        video = node.invoke(context)
    else:
        raise

Prevention

Feed only Wan denoiser outputs into the video node.
Avoid ad-hoc squeeze/reshape between denoiser and video node.
Log tensor shapes at workflow boundaries.
Use type-checked graph connections where possible.

When it happens

Trigger: Passing a tensor of ndim other than 4 or 5 into wan_latents_to_video — e.g. a 2D noise tensor, an image-latent [B,C,H,W] mis-shaped into 3D, or a corrupted tensor from an upstream node.

Common situations: Wiring an image latents node (4D) through reshaping that drops dims; feeding raw noise instead of denoiser output; version drift where an upstream node changed its output rank.

Related errors

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/699013383a95e378. Report an issue: GitHub.